John Berryman’s Post

Thanks for having us AI Rising. 2024 is the year of value realization for AI. Outcomes, not outputs and POCs. If you want to discuss pls DM or hit me up at mike@NeuZeitGroup.com.

This and all episodes at: https://aiandyou.net/ .   Artificial General Intelligence: Once upon a time, this was considered a pipe dream, a fantasy of dreamers with no sense of the practical limitations of real AI. That was last year. Now, AGI is an explicit goal of many enterprises, notably among them Simuli. Their CEO, Rachel St. Clair, co-founded the company with Ben Goertzel, who has also been on this show. Rachel is a Fellow of the Center for Future Mind, with a doctorate in Complex Systems and Brain Sciences from Florida Atlantic University. She researches artificial general intelligence, focusing on complex systems and neuromorphic learning algorithms. Her goal is to “help create human-like, conscious, artificial, general intelligence to help humans solve the worst of our problems.”  In part 1 we talk about markers for AGI, distinctions between it and narrow artificial intelligence, self-driving cars, robotics, and embodiment, and… disco balls. All this plus our usual look at today's AI headlines. Transcript and URLs referenced at HumanCusp Blog.          

There has been so much buzz about LLMs (mostly GPT-x) for educational applications. Perhaps, a more strategic and focused approach would be to leverage SLMs like phi-2 for key learning scenarios like reasoning, math, and reading. Added advantage - SLMs are so much more cost effective to run at scale in products - making them ideal for hyper cost-aware educational contexts. Check out this great conversation between Kevin Scott and Sebastien Bubeck to understand more! Well worth the time! #slm #phi2 #microsoftresearch #aiineducation

Not for general consumption; only for fellow AI enthusiasts... I know this is a couple of days old now, but it really is incredible from Anthropic. I remember doing a whole section of one of the very early The AI Exchange courses to manually set up this whole eval process. It was hugely valuable to learn the mechanics of how/why to do this manually, but jeepers it's amazing that a little over a year later this is now baked right into one of the premier AI tools, and so easy to operate for non-technical folks. The pace that this tech is moving at?! https://lnkd.in/gPxK-w49 

The latest edition in our series of short films on AI at the US Naval Academy is out. It's 16 minutes long, focuses on generative AI, and features Provost Samara Firebaugh's thoughts on where all of this is heading. https://lnkd.in/e69QH_HF  #artificialintelligence #generativeai

This year there have been an abundance of bears strolling on streets and trails of the North Shore. We posted a video about one - it was a cub, so naturally, we cautiously walked away while immortalizing the experience. Then THIS! 👇 Thanks Era Psodis for promptly flagging and helping us uncover the secret mysteries of the Ding Dong on all AI platforms! What is a Ding Dong? Please let us know in the comments section

In this week's episode of Scam of The Week, we're diving into the world of AI scams – a new trick in the cybercriminal's playbook. We'll break down how these AI scams work and how we can outsmart them. Watch here- https://lnkd.in/dc5yg2Rj

**Interview about Phi-2 and more** This new 12 minute interview between Sébastien Bubeck (leader of the Machine Learning Foundations group at Microsoft Research) and Microsoft CTO and EVP of AI, Kevin Scott, captures some interesting nuggets about how Phi models were trained with GPT-generated data. I especially like the analogies about teaching a child and creating a model with textbook-quality data. You wouldn't tech a child using all the garbage on the internet! Also very interesting to hear about the unique difficulties of bringing common sense reasoning to an SLM. The second half of the interview drifts into more esoteric topics, but still very interesting. https://lnkd.in/eMpEiQU8

I always enjoy these discussions with Chang. His depth of knowledge in the field and ability to explain it, makes for a good chat. If you'd like to find out more about multi-modal AI around the deep fake topic, this short video is a good place to start.

Meshing is often the trickiest part of #multiphysics #simulations. We’re working on something that could ease this pain with our upcoming AI-based meshing. Stay tuned and follow Quanscient for future updates!